Improving diagnosis and surveillance of childhood TB are key research priorities. We established intensified case finding and state of the art TB diagnostics to investigate the performance of clinical and laboratory tools for childhood TB diagnosis at 2 hospitals in Kenya. We estimated the community incidence of childhood TB using a continuous demographic surveillance survey and detailed surveillance sensitivity analysis.

2041 children were investigated for suspected TB. 70 (3.4%) had bacteriologically confirmed TB, 63 (3.1%) had clinically highly probable TB, and a further 144 (7.1%) were treated for TB based on their clinical presenting features. 107/133 (80%) confirmed/highly probable TB (CHPTB) cases had pulmonary TB. CHPTB was associated with HIV infection (OR 2.1, 95% CI 1.3-3.2), malnutrition (1.5, 1.0-2.1) and close TB contact (5.7, 3.8-8.5). The population attributable fraction of a known close TB contact was 38.8-52.5%. The estimated community incidence of CHPTB locally and nationally was 46 and 83 per 100,000 per year, respectively.

The performance of published clinical diagnostic tools varied widely, but the accuracy of all was limited. We derived and independently validated a simple KIDS TB Score that ruled out TB in 2/3 suspects with 98.8% negative predictive value, stratifying other children into groups of increasing risk. Bacteriological yield was highest for the Mycobacterial Growth Inhibitor Tube (MGIT) method (sensitivity 34%, 29-39%, among CHPTB patient samples), and lower for the Microscopic Observation Drug Susceptibility (MODS) assay (30%, 24-35%).

The study provides the first comprehensive description from the region of the clinical spectrum of childhood TB, and the only prospective incidence estimates. It suggests up to half of all cases are potentially preventable by implementing current recommendations for isoniazid chemoprophylaxis. The diagnostic performance of clinical and laboratory methods should inform development of future clinical guidelines and laboratory capacity.